2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/spinlock.h>
11 #include <linux/completion.h>
12 #include <linux/buffer_head.h>
13 #include <linux/gfs2_ondisk.h>
14 #include <linux/crc32.h>
28 #include "trace_gfs2.h"
30 /* This doesn't need to be that large as max 64 bit pointers in a 4k
31 * block is 512, so __u16 is fine for that. It saves stack space to
35 struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
36 __u16 mp_list[GFS2_MAX_META_HEIGHT];
39 typedef int (*block_call_t) (struct gfs2_inode *ip, struct buffer_head *dibh,
40 struct buffer_head *bh, __be64 *top,
41 __be64 *bottom, unsigned int height,
46 unsigned int sm_height;
50 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
52 * @dibh: the dinode buffer
53 * @block: the block number that was allocated
54 * @page: The (optional) page. This is looked up if @page is NULL
59 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
60 u64 block, struct page *page)
62 struct inode *inode = &ip->i_inode;
63 struct buffer_head *bh;
66 if (!page || page->index) {
67 page = grab_cache_page(inode->i_mapping, 0);
73 if (!PageUptodate(page)) {
74 void *kaddr = kmap(page);
75 u64 dsize = i_size_read(inode);
77 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
78 dsize = dibh->b_size - sizeof(struct gfs2_dinode);
80 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
81 memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
84 SetPageUptodate(page);
87 if (!page_has_buffers(page))
88 create_empty_buffers(page, 1 << inode->i_blkbits,
91 bh = page_buffers(page);
93 if (!buffer_mapped(bh))
94 map_bh(bh, inode->i_sb, block);
96 set_buffer_uptodate(bh);
97 if (!gfs2_is_jdata(ip))
98 mark_buffer_dirty(bh);
99 if (!gfs2_is_writeback(ip))
100 gfs2_trans_add_bh(ip->i_gl, bh, 0);
104 page_cache_release(page);
111 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
112 * @ip: The GFS2 inode to unstuff
113 * @page: The (optional) page. This is looked up if the @page is NULL
115 * This routine unstuffs a dinode and returns it to a "normal" state such
116 * that the height can be grown in the traditional way.
121 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
123 struct buffer_head *bh, *dibh;
124 struct gfs2_dinode *di;
126 int isdir = gfs2_is_dir(ip);
129 down_write(&ip->i_rw_mutex);
131 error = gfs2_meta_inode_buffer(ip, &dibh);
135 if (i_size_read(&ip->i_inode)) {
136 /* Get a free block, fill it with the stuffed data,
137 and write it out to disk */
140 error = gfs2_alloc_block(ip, &block, &n);
144 gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
145 error = gfs2_dir_get_new_buffer(ip, block, &bh);
148 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
149 dibh, sizeof(struct gfs2_dinode));
152 error = gfs2_unstuffer_page(ip, dibh, block, page);
158 /* Set up the pointer to the new block */
160 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
161 di = (struct gfs2_dinode *)dibh->b_data;
162 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
164 if (i_size_read(&ip->i_inode)) {
165 *(__be64 *)(di + 1) = cpu_to_be64(block);
166 gfs2_add_inode_blocks(&ip->i_inode, 1);
167 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
171 di->di_height = cpu_to_be16(1);
176 up_write(&ip->i_rw_mutex);
182 * find_metapath - Find path through the metadata tree
183 * @sdp: The superblock
184 * @mp: The metapath to return the result in
185 * @block: The disk block to look up
186 * @height: The pre-calculated height of the metadata tree
188 * This routine returns a struct metapath structure that defines a path
189 * through the metadata of inode "ip" to get to block "block".
192 * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
193 * filesystem with a blocksize of 4096.
195 * find_metapath() would return a struct metapath structure set to:
196 * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
197 * and mp_list[2] = 165.
199 * That means that in order to get to the block containing the byte at
200 * offset 101342453, we would load the indirect block pointed to by pointer
201 * 0 in the dinode. We would then load the indirect block pointed to by
202 * pointer 48 in that indirect block. We would then load the data block
203 * pointed to by pointer 165 in that indirect block.
205 * ----------------------------------------
210 * ----------------------------------------
214 * ----------------------------------------
218 * |0 5 6 7 8 9 0 1 2|
219 * ----------------------------------------
223 * ----------------------------------------
228 * ----------------------------------------
232 * ----------------------------------------
233 * | Data block containing offset |
237 * ----------------------------------------
241 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
242 struct metapath *mp, unsigned int height)
246 for (i = height; i--;)
247 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
251 static inline unsigned int metapath_branch_start(const struct metapath *mp)
253 if (mp->mp_list[0] == 0)
259 * metapointer - Return pointer to start of metadata in a buffer
260 * @height: The metadata height (0 = dinode)
263 * Return a pointer to the block number of the next height of the metadata
264 * tree given a buffer containing the pointer to the current height of the
268 static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
270 struct buffer_head *bh = mp->mp_bh[height];
271 unsigned int head_size = (height > 0) ?
272 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
273 return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
277 * lookup_metapath - Walk the metadata tree to a specific point
281 * Assumes that the inode's buffer has already been looked up and
282 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
283 * by find_metapath().
285 * If this function encounters part of the tree which has not been
286 * allocated, it returns the current height of the tree at the point
287 * at which it found the unallocated block. Blocks which are found are
288 * added to the mp->mp_bh[] list.
290 * Returns: error or height of metadata tree
293 static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
295 unsigned int end_of_metadata = ip->i_height - 1;
301 for (x = 0; x < end_of_metadata; x++) {
302 ptr = metapointer(x, mp);
303 dblock = be64_to_cpu(*ptr);
307 ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, 0, &mp->mp_bh[x+1]);
315 static inline void release_metapath(struct metapath *mp)
319 for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
320 if (mp->mp_bh[i] == NULL)
322 brelse(mp->mp_bh[i]);
327 * gfs2_extent_length - Returns length of an extent of blocks
328 * @start: Start of the buffer
329 * @len: Length of the buffer in bytes
330 * @ptr: Current position in the buffer
331 * @limit: Max extent length to return (0 = unlimited)
332 * @eob: Set to 1 if we hit "end of block"
334 * If the first block is zero (unallocated) it will return the number of
335 * unallocated blocks in the extent, otherwise it will return the number
336 * of contiguous blocks in the extent.
338 * Returns: The length of the extent (minimum of one block)
341 static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob)
343 const __be64 *end = (start + len);
344 const __be64 *first = ptr;
345 u64 d = be64_to_cpu(*ptr);
352 if (limit && --limit == 0)
356 } while(be64_to_cpu(*ptr) == d);
359 return (ptr - first);
362 static inline void bmap_lock(struct gfs2_inode *ip, int create)
365 down_write(&ip->i_rw_mutex);
367 down_read(&ip->i_rw_mutex);
370 static inline void bmap_unlock(struct gfs2_inode *ip, int create)
373 up_write(&ip->i_rw_mutex);
375 up_read(&ip->i_rw_mutex);
378 static inline __be64 *gfs2_indirect_init(struct metapath *mp,
379 struct gfs2_glock *gl, unsigned int i,
380 unsigned offset, u64 bn)
382 __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
383 ((i > 1) ? sizeof(struct gfs2_meta_header) :
384 sizeof(struct gfs2_dinode)));
386 BUG_ON(mp->mp_bh[i] != NULL);
387 mp->mp_bh[i] = gfs2_meta_new(gl, bn);
388 gfs2_trans_add_bh(gl, mp->mp_bh[i], 1);
389 gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
390 gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
392 *ptr = cpu_to_be64(bn);
398 ALLOC_GROW_DEPTH = 1,
399 ALLOC_GROW_HEIGHT = 2,
400 /* ALLOC_UNSTUFF = 3, TBD and rather complicated */
404 * gfs2_bmap_alloc - Build a metadata tree of the requested height
405 * @inode: The GFS2 inode
406 * @lblock: The logical starting block of the extent
407 * @bh_map: This is used to return the mapping details
409 * @sheight: The starting height (i.e. whats already mapped)
410 * @height: The height to build to
411 * @maxlen: The max number of data blocks to alloc
413 * In this routine we may have to alloc:
414 * i) Indirect blocks to grow the metadata tree height
415 * ii) Indirect blocks to fill in lower part of the metadata tree
418 * The function is in two parts. The first part works out the total
419 * number of blocks which we need. The second part does the actual
420 * allocation asking for an extent at a time (if enough contiguous free
421 * blocks are available, there will only be one request per bmap call)
422 * and uses the state machine to initialise the blocks in order.
424 * Returns: errno on error
427 static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
428 struct buffer_head *bh_map, struct metapath *mp,
429 const unsigned int sheight,
430 const unsigned int height,
431 const unsigned int maxlen)
433 struct gfs2_inode *ip = GFS2_I(inode);
434 struct gfs2_sbd *sdp = GFS2_SB(inode);
435 struct buffer_head *dibh = mp->mp_bh[0];
437 unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
439 unsigned ptrs_per_blk;
440 const unsigned end_of_metadata = height - 1;
442 enum alloc_state state;
447 BUG_ON(dibh == NULL);
449 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
451 if (height == sheight) {
452 struct buffer_head *bh;
453 /* Bottom indirect block exists, find unalloced extent size */
454 ptr = metapointer(end_of_metadata, mp);
455 bh = mp->mp_bh[end_of_metadata];
456 dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
461 /* Need to allocate indirect blocks */
462 ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
463 dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]);
464 if (height == ip->i_height) {
465 /* Writing into existing tree, extend tree down */
466 iblks = height - sheight;
467 state = ALLOC_GROW_DEPTH;
469 /* Building up tree height */
470 state = ALLOC_GROW_HEIGHT;
471 iblks = height - ip->i_height;
472 branch_start = metapath_branch_start(mp);
473 iblks += (height - branch_start);
477 /* start of the second part of the function (state machine) */
479 blks = dblks + iblks;
484 error = gfs2_alloc_block(ip, &bn, &n);
488 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
489 gfs2_trans_add_unrevoke(sdp, bn, n);
491 /* Growing height of tree */
492 case ALLOC_GROW_HEIGHT:
494 ptr = (__be64 *)(dibh->b_data +
495 sizeof(struct gfs2_dinode));
498 for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
499 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
500 if (i - 1 == height - ip->i_height) {
502 gfs2_buffer_copy_tail(mp->mp_bh[i],
503 sizeof(struct gfs2_meta_header),
504 dibh, sizeof(struct gfs2_dinode));
505 gfs2_buffer_clear_tail(dibh,
506 sizeof(struct gfs2_dinode) +
508 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
509 sizeof(struct gfs2_meta_header));
511 state = ALLOC_GROW_DEPTH;
512 for(i = branch_start; i < height; i++) {
513 if (mp->mp_bh[i] == NULL)
515 brelse(mp->mp_bh[i]);
522 /* Branching from existing tree */
523 case ALLOC_GROW_DEPTH:
524 if (i > 1 && i < height)
525 gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1);
526 for (; i < height && n > 0; i++, n--)
527 gfs2_indirect_init(mp, ip->i_gl, i,
528 mp->mp_list[i-1], bn++);
533 /* Tree complete, adding data blocks */
536 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
537 gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1);
539 ptr = metapointer(end_of_metadata, mp);
542 *ptr++ = cpu_to_be64(bn++);
545 } while ((state != ALLOC_DATA) || !dblock);
547 ip->i_height = height;
548 gfs2_add_inode_blocks(&ip->i_inode, alloced);
549 gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
550 map_bh(bh_map, inode->i_sb, dblock);
551 bh_map->b_size = dblks << inode->i_blkbits;
552 set_buffer_new(bh_map);
557 * gfs2_block_map - Map a block from an inode to a disk block
559 * @lblock: The logical block number
560 * @bh_map: The bh to be mapped
561 * @create: True if its ok to alloc blocks to satify the request
563 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
564 * read of metadata will be required before the next block can be
565 * mapped. Sets buffer_new() if new blocks were allocated.
570 int gfs2_block_map(struct inode *inode, sector_t lblock,
571 struct buffer_head *bh_map, int create)
573 struct gfs2_inode *ip = GFS2_I(inode);
574 struct gfs2_sbd *sdp = GFS2_SB(inode);
575 unsigned int bsize = sdp->sd_sb.sb_bsize;
576 const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
577 const u64 *arr = sdp->sd_heightsize;
584 struct buffer_head *bh;
589 memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
590 bmap_lock(ip, create);
591 clear_buffer_mapped(bh_map);
592 clear_buffer_new(bh_map);
593 clear_buffer_boundary(bh_map);
594 trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
595 if (gfs2_is_dir(ip)) {
596 bsize = sdp->sd_jbsize;
597 arr = sdp->sd_jheightsize;
600 ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
604 height = ip->i_height;
605 size = (lblock + 1) * bsize;
606 while (size > arr[height])
608 find_metapath(sdp, lblock, &mp, height);
610 if (height > ip->i_height || gfs2_is_stuffed(ip))
612 ret = lookup_metapath(ip, &mp);
615 if (ret != ip->i_height)
617 ptr = metapointer(ip->i_height - 1, &mp);
620 map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
621 bh = mp.mp_bh[ip->i_height - 1];
622 len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
623 bh_map->b_size = (len << inode->i_blkbits);
625 set_buffer_boundary(bh_map);
628 release_metapath(&mp);
629 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
630 bmap_unlock(ip, create);
634 /* All allocations are done here, firstly check create flag */
636 BUG_ON(gfs2_is_stuffed(ip));
641 /* At this point ret is the tree depth of already allocated blocks */
642 ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
647 * Deprecated: do not use in new code
649 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
651 struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
659 bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
660 ret = gfs2_block_map(inode, lblock, &bh, create);
661 *extlen = bh.b_size >> inode->i_blkbits;
662 *dblock = bh.b_blocknr;
671 * recursive_scan - recursively scan through the end of a file
673 * @dibh: the dinode buffer
674 * @mp: the path through the metadata to the point to start
675 * @height: the height the recursion is at
676 * @block: the indirect block to look at
677 * @first: 1 if this is the first block
678 * @bc: the call to make for each piece of metadata
679 * @data: data opaque to this function to pass to @bc
681 * When this is first called @height and @block should be zero and
682 * @first should be 1.
687 static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
688 struct metapath *mp, unsigned int height,
689 u64 block, int first, block_call_t bc,
692 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
693 struct buffer_head *bh = NULL;
694 __be64 *top, *bottom;
697 int mh_size = sizeof(struct gfs2_meta_header);
700 error = gfs2_meta_inode_buffer(ip, &bh);
705 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
706 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
708 error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh);
712 top = (__be64 *)(bh->b_data + mh_size) +
713 (first ? mp->mp_list[height] : 0);
715 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
718 error = bc(ip, dibh, bh, top, bottom, height, data);
722 if (height < ip->i_height - 1)
723 for (; top < bottom; top++, first = 0) {
727 bn = be64_to_cpu(*top);
729 error = recursive_scan(ip, dibh, mp, height + 1, bn,
741 * do_strip - Look for a layer a particular layer of the file and strip it off
743 * @dibh: the dinode buffer
744 * @bh: A buffer of pointers
745 * @top: The first pointer in the buffer
746 * @bottom: One more than the last pointer
747 * @height: the height this buffer is at
748 * @data: a pointer to a struct strip_mine
753 static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
754 struct buffer_head *bh, __be64 *top, __be64 *bottom,
755 unsigned int height, void *data)
757 struct strip_mine *sm = data;
758 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
759 struct gfs2_rgrp_list rlist;
763 unsigned int rg_blocks = 0;
765 unsigned int revokes = 0;
772 if (height != sm->sm_height)
780 metadata = (height != ip->i_height - 1);
782 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
783 else if (ip->i_depth)
784 revokes = sdp->sd_inptrs;
786 if (ip != GFS2_I(sdp->sd_rindex))
787 error = gfs2_rindex_hold(sdp, &ip->i_alloc->al_ri_gh);
788 else if (!sdp->sd_rgrps)
789 error = gfs2_ri_update(ip);
794 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
798 for (p = top; p < bottom; p++) {
802 bn = be64_to_cpu(*p);
804 if (bstart + blen == bn)
808 gfs2_rlist_add(sdp, &rlist, bstart);
816 gfs2_rlist_add(sdp, &rlist, bstart);
818 goto out; /* Nothing to do */
820 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
822 for (x = 0; x < rlist.rl_rgrps; x++) {
823 struct gfs2_rgrpd *rgd;
824 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
825 rg_blocks += rgd->rd_length;
828 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
832 error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
833 RES_INDIRECT + RES_STATFS + RES_QUOTA,
838 down_write(&ip->i_rw_mutex);
840 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
841 gfs2_trans_add_bh(ip->i_gl, bh, 1);
847 for (p = top; p < bottom; p++) {
851 bn = be64_to_cpu(*p);
853 if (bstart + blen == bn)
858 __gfs2_free_meta(ip, bstart, blen);
860 __gfs2_free_data(ip, bstart, blen);
870 gfs2_add_inode_blocks(&ip->i_inode, -1);
874 __gfs2_free_meta(ip, bstart, blen);
876 __gfs2_free_data(ip, bstart, blen);
881 gfs2_statfs_change(sdp, 0, +btotal, 0);
882 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
885 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
887 gfs2_dinode_out(ip, dibh->b_data);
889 up_write(&ip->i_rw_mutex);
894 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
896 gfs2_rlist_free(&rlist);
898 if (ip != GFS2_I(sdp->sd_rindex))
899 gfs2_glock_dq_uninit(&ip->i_alloc->al_ri_gh);
904 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
906 * This is partly borrowed from ext3.
908 static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
910 struct inode *inode = mapping->host;
911 struct gfs2_inode *ip = GFS2_I(inode);
912 unsigned long index = from >> PAGE_CACHE_SHIFT;
913 unsigned offset = from & (PAGE_CACHE_SIZE-1);
914 unsigned blocksize, iblock, length, pos;
915 struct buffer_head *bh;
919 page = grab_cache_page(mapping, index);
923 blocksize = inode->i_sb->s_blocksize;
924 length = blocksize - (offset & (blocksize - 1));
925 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
927 if (!page_has_buffers(page))
928 create_empty_buffers(page, blocksize, 0);
930 /* Find the buffer that contains "offset" */
931 bh = page_buffers(page);
933 while (offset >= pos) {
934 bh = bh->b_this_page;
941 if (!buffer_mapped(bh)) {
942 gfs2_block_map(inode, iblock, bh, 0);
943 /* unmapped? It's a hole - nothing to do */
944 if (!buffer_mapped(bh))
948 /* Ok, it's mapped. Make sure it's up-to-date */
949 if (PageUptodate(page))
950 set_buffer_uptodate(bh);
952 if (!buffer_uptodate(bh)) {
954 ll_rw_block(READ, 1, &bh);
956 /* Uhhuh. Read error. Complain and punt. */
957 if (!buffer_uptodate(bh))
962 if (!gfs2_is_writeback(ip))
963 gfs2_trans_add_bh(ip->i_gl, bh, 0);
965 zero_user(page, offset, length);
966 mark_buffer_dirty(bh);
969 page_cache_release(page);
973 static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
975 struct gfs2_inode *ip = GFS2_I(inode);
976 struct gfs2_sbd *sdp = GFS2_SB(inode);
977 struct address_space *mapping = inode->i_mapping;
978 struct buffer_head *dibh;
979 int journaled = gfs2_is_jdata(ip);
982 error = gfs2_trans_begin(sdp,
983 RES_DINODE + (journaled ? RES_JDATA : 0), 0);
987 error = gfs2_meta_inode_buffer(ip, &dibh);
991 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
993 if (gfs2_is_stuffed(ip)) {
994 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
996 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
997 error = gfs2_block_truncate_page(mapping, newsize);
1001 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1004 i_size_write(inode, newsize);
1005 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1006 gfs2_dinode_out(ip, dibh->b_data);
1008 truncate_pagecache(inode, oldsize, newsize);
1012 gfs2_trans_end(sdp);
1016 static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
1018 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1019 unsigned int height = ip->i_height;
1027 lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
1029 find_metapath(sdp, lblock, &mp, ip->i_height);
1030 if (!gfs2_alloc_get(ip))
1033 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1038 struct strip_mine sm;
1039 sm.sm_first = !!size;
1040 sm.sm_height = height;
1042 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, do_strip, &sm);
1047 gfs2_quota_unhold(ip);
1054 static int trunc_end(struct gfs2_inode *ip)
1056 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1057 struct buffer_head *dibh;
1060 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1064 down_write(&ip->i_rw_mutex);
1066 error = gfs2_meta_inode_buffer(ip, &dibh);
1070 if (!i_size_read(&ip->i_inode)) {
1072 ip->i_goal = ip->i_no_addr;
1073 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1075 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1076 ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1078 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1079 gfs2_dinode_out(ip, dibh->b_data);
1083 up_write(&ip->i_rw_mutex);
1084 gfs2_trans_end(sdp);
1089 * do_shrink - make a file smaller
1091 * @oldsize: the current inode size
1092 * @newsize: the size to make the file
1094 * Called with an exclusive lock on @inode. The @size must
1095 * be equal to or smaller than the current inode size.
1100 static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1102 struct gfs2_inode *ip = GFS2_I(inode);
1105 error = trunc_start(inode, oldsize, newsize);
1108 if (gfs2_is_stuffed(ip))
1111 error = trunc_dealloc(ip, newsize);
1113 error = trunc_end(ip);
1118 void gfs2_trim_blocks(struct inode *inode)
1120 u64 size = inode->i_size;
1123 ret = do_shrink(inode, size, size);
1128 * do_grow - Touch and update inode size
1130 * @size: The new size
1132 * This function updates the timestamps on the inode and
1133 * may also increase the size of the inode. This function
1134 * must not be called with @size any smaller than the current
1137 * Although it is not strictly required to unstuff files here,
1138 * earlier versions of GFS2 have a bug in the stuffed file reading
1139 * code which will result in a buffer overrun if the size is larger
1140 * than the max stuffed file size. In order to prevent this from
1141 * occurring, such files are unstuffed, but in other cases we can
1142 * just update the inode size directly.
1144 * Returns: 0 on success, or -ve on error
1147 static int do_grow(struct inode *inode, u64 size)
1149 struct gfs2_inode *ip = GFS2_I(inode);
1150 struct gfs2_sbd *sdp = GFS2_SB(inode);
1151 struct buffer_head *dibh;
1152 struct gfs2_alloc *al = NULL;
1155 if (gfs2_is_stuffed(ip) &&
1156 (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1157 al = gfs2_alloc_get(ip);
1161 error = gfs2_quota_lock_check(ip);
1163 goto do_grow_alloc_put;
1165 al->al_requested = 1;
1166 error = gfs2_inplace_reserve(ip);
1168 goto do_grow_qunlock;
1171 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0);
1173 goto do_grow_release;
1176 error = gfs2_unstuff_dinode(ip, NULL);
1181 error = gfs2_meta_inode_buffer(ip, &dibh);
1185 i_size_write(inode, size);
1186 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1187 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1188 gfs2_dinode_out(ip, dibh->b_data);
1192 gfs2_trans_end(sdp);
1195 gfs2_inplace_release(ip);
1197 gfs2_quota_unlock(ip);
1205 * gfs2_setattr_size - make a file a given size
1207 * @newsize: the size to make the file
1209 * The file size can grow, shrink, or stay the same size. This
1210 * is called holding i_mutex and an exclusive glock on the inode
1216 int gfs2_setattr_size(struct inode *inode, u64 newsize)
1221 BUG_ON(!S_ISREG(inode->i_mode));
1223 ret = inode_newsize_ok(inode, newsize);
1227 oldsize = inode->i_size;
1228 if (newsize >= oldsize)
1229 return do_grow(inode, newsize);
1231 return do_shrink(inode, oldsize, newsize);
1234 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1237 error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1239 error = trunc_end(ip);
1243 int gfs2_file_dealloc(struct gfs2_inode *ip)
1245 return trunc_dealloc(ip, 0);
1249 * gfs2_write_alloc_required - figure out if a write will require an allocation
1250 * @ip: the file being written to
1251 * @offset: the offset to write to
1252 * @len: the number of bytes being written
1254 * Returns: 1 if an alloc is required, 0 otherwise
1257 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1260 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1261 struct buffer_head bh;
1263 u64 lblock, lblock_stop, size;
1269 if (gfs2_is_stuffed(ip)) {
1271 sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1276 shift = sdp->sd_sb.sb_bsize_shift;
1277 BUG_ON(gfs2_is_dir(ip));
1278 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1279 lblock = offset >> shift;
1280 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1281 if (lblock_stop > end_of_file)
1284 size = (lblock_stop - lblock) << shift;
1288 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1289 if (!buffer_mapped(&bh))
1292 lblock += (bh.b_size >> ip->i_inode.i_blkbits);